Bill Gates, Hillary Clinton, and a cluster of AI company executives have all sounded alarms about the rate of change in artificial intelligence development. They don’t all agree on what causes them the most concern, but there is a common refrain that effectively says, “We don’t know how to stop the technology if it gets out of hand.”
■ That isn’t an unwarranted concern. The language of “AI could kill us all” seems a bit alarmist, but any truly existential risk ought to be taken seriously. Even a low-probability risk deserves attention when the stakes are extremely high.
■ But we shouldn’t let the sci-fi concern overshadow lower-stakes risks that are utterly certain. It is absolutely the case that artificial intelligence tools are being deployed in coordinated ways to cause lots of social-scale troubles. Misinformation and disinformation are already being generated at scale, and it is completely certain that bad actors with lots of bots to deploy are going to use them to create cunning and manipulative schemes to steal and defraud people by the millions.
■ On the personal level, the jeopardy could be even worse. It should be easy to imagine malicious people with axes to grind using AI tools to carry out terrible vendettas against others: Romantic ex-partners, business competitors, electoral rivals...the list could be endless. All it really takes is for an unscrupulous person to assemble the right prompt to get Grok or Gemini or Claude or any of the rest to build them customized action plans to ruin someone else’s reputation, business, credit score, or other store of social value. Some of the tools would even execute the plan, for the right combination of tokens and prompts.
■ It’s not enough to say that we should worry about far-out, hard-to-fathom misuses of a frontier technology when there are simple, almost pedestrian ways in which the tools can be used for evil on a much smaller scale. Worrying about a technology out of control isn’t wrong -- but we can’t overlook the plain fact that we don’t need self-interested, out-of-control machines for harm to be done.
■ Self-interested, out-of-control human beings are already here, and the harms they can conjure up when armed with extremely powerful computing tools are already more than enough to be worthy of attention. A tool that could potentially destroy every life demands close attention, but a tool that could be used to ruin one life at a time (but in volume) could be nearly as bad.